dsh-agent-doctor

After mounting tools in a DeepSeek Harness, developers often struggle to quickly confirm whether the effective tool surface exposed to the model matches their expectations. Which tools are visible, whether reserved capabilities have accidentally leaked, and if any tool name looks side-effect-shaped are details that are tedious and error-prone to verify by hand. dsh-agent-doctor provides a /doctor command that lets you run a bounded health check on the model-visible tool surface from an interactive DSH interface.

The diagnostics cover five dimensions: listing the count and names of effective tools, detecting whether the reserved run_code capability is unexpectedly visible, issuing a heuristic warning on tool names that look side-effect-shaped, verifying that an AgentFuse policy boundary is present as required, and reporting an approval-service availability signal. The name-based check is a diagnostic hint only; the plugin neither authorizes nor blocks execution and does not act as a security scanner or enforcement boundary. True runtime enforcement should still be delegated to a policy layer such as AgentFuse.

Ideal for developers building or debugging DSH agents. If you need a quick pre-release health check on the tool surface, want to confirm that reserved capabilities have not been unintentionally exposed, or simply want a structured configuration snapshot in your logs, this lightweight plugin fits the bill. The core checker has no runtime dependencies and can be tested in isolation.